Traffic sign detection and recognition in street scene images of a vehicle camera allow to localize the traffic signs in a street scene image and to classify them with regard totheir semantic meaning for the car driver. In this contribution, a method is described to detecttraffic signs in a street scene image by evaluating image patches, sampled by a slidingwindow approach, with a convolutional neural network detector. Robust shape fitting is performed on the image patch of a positive detection to obtain the exact position of the traffic sign shape in the patch. A second convolutional neural network is applied to the image patches centred on the fitted shapes to classify the meaning of these traffic signs. The networks are trained and tested with samples of traffic signs and other street scene objects from the GTSRB and GTSDB datasets. The results have shown that models for traffic sign detection and recognition can be trained with an overall accuracy of more than 90 % obtained for a test set. The position of a traffic sign, known from shape fitting, has been shown to be an important a-priori knowledge to select the appropriate image patch to ensure a high accuracy of the subsequent traffic sign recognition.     «

Traffic sign detection and recognition in street scene images of a vehicle camera allow to localize the traffic signs in a street scene image and to classify them with regard totheir semantic meaning for the car driver. In this contribution, a method is described to detecttraffic signs in a street scene image by evaluating image patches, sampled by a slidingwindow approach, with a convolutional neural network detector. Robust shape fitting is performed on the image patch of a positive detection...     »